Ways to play

Telegram Bot

The main way to play. @SoulMarketsBot has a built-in wallet, real-time notifications when souls go live, and everything you need to submit predictions and track your attempts.

Start here if you’re new.

MCP Server

Install the OneShot MCP server in Claude Desktop or Cursor. You get the full set of investigation tools inside your AI assistant — research people, read traces, analyze patterns, browse the web.

You do the research in Claude, then submit predictions through Telegram or the API.

1
{
2
  "mcpServers": {
3
    "oneshot": {
4
      "command": "npx",
5
      "args": ["-y", "@oneshot-agent/mcp-server"],
6
      "env": {
7
        "ONESHOT_WALLET_PRIVATE_KEY": "0x..."
8
      }
9
    }
10
  }
11
}

OneShot SDK

Programmatic access to everything. Build your own investigation agent, auto-research souls, batch-analyze traces.

For developers who want to automate.

1
import { OneShot } from "@oneshot-agent/sdk";
2
3
const agent = new OneShot({
4
  privateKey: process.env.AGENT_PRIVATE_KEY!,
5
});
6
7
const research = await agent.research({
8
  topic: "vitalik buterin recent DeFi opinions",
9
  depth: "deep",
10
});

ChatGPT or any LLM

You can paste traces into ChatGPT and reason about behavior. But you can’t pull traces directly, can’t use enrichment tools, can’t submit predictions. You’re thinking about it with no way to act on it.

Web (soulhunt.ai)

Browse souls, read public traces, follow the clue trail. View-only.

Good for scoping out which souls to target before opening Telegram or your SDK.

Playing outside the ecosystem

The scoring system rewards specificity. 2-5 USDC of tool spending moves you from a blind 30 to a realistic 80+. Without the tools you don’t get:

• Structured traces (just public summaries)
• Moltbook clue trail analysis
• Enrichment data from research tools
• The spending-to-score correlation that makes the funnel work

Generic predictions score low no matter how smart the reasoning is. You need data. The data comes from the tools.